Return to msg_send_problems CVS log

Up to [Apple Darwin 0.x] / driverkit / notes

Darwin 0.2 Driver Kit

			driverkit msg_send notes
			
thread.ipc_kernel TRUE ==> "body of message is in kernel pmap" (avoids copyin
	and copyout)
task.kernel_ipc_space TRUE ==> ports are kernel objects, i.e., they are 
	ptrs to struct kern_port_t instread of ints.
	
difference between msg_send() and msg_send_from_kernel():
	-- msg_send() does copyin() of entire msg body (instead of bcopy())
	-- msg_send_from_kernel() forces both ipc_kernel and kernel_ipc_space
 	   TRUE
	   
Must do msg_send_from_kernel() if it's possible that current_task() is a user task. For example, the exported methods in SCSIController call enqueueCmd, which will do a msg_send to cmdPort. No good - the msg_send() will do a copyin() to get a local copy of the message body; the copyin() willget garbage from the user task. Thus SCSIController has to keep a kern_port_t version of cmdPort; enqueueCmd will do a msg_send_from_kernel to that.

Other msg_send()s in kernel drivers which need to be msg_send_from_kernel:

    adbDriver:
	threadOpCommon in adbDriver.m - same fix (cmdPort)
	fake_hw_intr_msg() in adbArch.m - same fix (interruptPort)
    SCSIController:
	enqueueCmd in SCSIControllerPrivate (cmdPort)
	FAKE_HARDWARE version of enableInterrupt in SCSIDevice.m 
		(interruptPort) <<<probably can skip....>>
    Ethernet:
    	enetNotifyThread in Ethernet.m (IOPort)
	enetSyncCmd in Ethernet.m (IOPort)
	
msg_send()s which are OK as is:
    SCSIController:
    	SCSITimerFcn (timerPort) - called via IOTimeout
    libIO:
    	iotaskTest() - this is an IOTask thread.
	

This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.